BNL Ninja OS
Custom Linux for AI servers: NVIDIA drivers, CUDA and NCCL pinned per release, A/B rootfs with automatic rollback, read-only root where the workload allows, and a first-class offline install path for air-gapped sites.
Hardware, OS, console, client
One engagement covers the whole path: right-sized AI hardware, BNL Ninja OS, Ninja Console on your server, and the BNL Ninja Client on every device. Installed, supported, yours.
Every Ninja Node ships with the same software foundation; only the silicon scales.
Custom Linux for AI servers: NVIDIA drivers, CUDA and NCCL pinned per release, A/B rootfs with automatic rollback, read-only root where the workload allows, and a first-class offline install path for air-gapped sites.
The web control plane on your server: deploy and swap models, manage users with RBAC, wire in your identity provider, manage knowledge bases and RAG, launch fine-tune jobs, and read the audit trail.
One client on Windows, macOS, Linux, iOS, and Android: chat, coding, tool calling, computer use, RAG over your knowledge bases, and connectivity to your internal tools — all talking to your server, never a third-party cloud.
Reference configurations — final hardware is quoted per engagement against your models and concurrency.
| Tier | Users | Reference configuration | GPU memory |
|---|---|---|---|
| Ninja Node N1 | 1–5 | Single-box workstation, 1× GPU (RTX 6000 Ada / RTX 5090 class) | 24–48 GB |
| Ninja Node N2 | 6–25 | 1U/2U single node, 2–4 GPUs (L4/L20/RTX 6000 Ada class) | 48–192 GB |
| Ninja Node N3 | 26–100 | 8-GPU HGX-class node (H100/H200 or MI300X class) | 640–1,128 GB |
| Ninja Node N4 | 100–1,000 | Enterprise Intelligence Infrastructure: multi-node cluster, NVLink/InfiniBand fabric for board-level analytics and sovereign corporate intelligence | Scaled to model + concurrency |
Reference configurations for sizing conversations. Final hardware is quoted per engagement.
The Ninja Console and Client share one API surface on your server. There is an OpenAI-compatible endpoint for chat and embeddings, so your existing tools plug straight in, plus BNL-proprietary endpoints for RAG, knowledge bases, fine-tuning, RBAC, identity and audit. You keep the freedom to bring any tool. You never hand your data to anyone.
Tell us your user counts and which models. We will recommend a tier and what it costs to own it.
Request a sizing